Growth Accelerations and Regime Changes: A Correction
Richard JongAPin^{a }and Jakob de Haan^{a,b}
^{a} Faculty of Economics and Business, University of Groningen, The Netherlands
^{b} CESifo Munich, Germany
Version, 22 September 2007
Abstract
We argue that the finding of Hausmann et al. (2005) that a political regime change increases the probability of an economic growth acceleration is wrong and the result of an error in their database. When we correct for this error and stick to the definition of regime change of Hausmann et al., we find that regime changes do not affect the likelihood that a growth acceleration occurs. We also find some evidence that economic liberalization increases the probability of a growth acceleration, independent of whether this acceleration is sustained.
Key words: economic growth, growth accelerations, regime changes
JEL code: O17, O11
We like to thank Dani Rodrik for providing the data used in the analysis.
Corresponding author: Jakob de Haan, Faculty of Economics and Business, University of Groningen, PO Box 800, 9700 AV Groningen, The Netherlands, Tel. 31(0)503633706; Fax 31(0)503633720; email: jakob.de.haan@rug.nl.
“Economists treat replication the way teenagers treat chastity—as an ideal to be professed but not to be practiced” (Hamermesh, 2007, p.1).
1. Introduction
There is much research on the impact of political, legal, and economic institutions on long term economic growth. However, the usefulness of the growth regression framework is questionable as it assumes that a single linear model is appropriate for all countries at all times (De Haan, 2007). Very few countries have experienced consistently constant growth rates over time. Pritchett (2000) documents, for instance, that the variation in growth rates within countries is large relative to both the average growth rates as well as the variance across countries. Likewise, Jones and Olken (2005) report that no less than 48 countries have experienced one or more structural breaks in their economic development. These breaks lead to very distinct growth patterns. Whereas some countries have experienced long periods of sustained growth, others faced rapid growth followed by stagnation or even a period of crisis. Still, other countries face continuous stagnation or steady decline. Consequently, empirical growth research has underestimated the importance of instability and volatility in growth rates, especially in developing countries.
One promising research strategy is to examine the economic, political, institutional and policy conditions that accompany changes in growth patterns. A pioneering contribution in this field is by Hausmann, Pritchett and Rodrik (2005) who examine whether political regime changes and economic reforms precede growth accelerations. Hausmann et al. (2005) identify more than 80 growth accelerations since the 1950s, which tend to be highly unpredictable. They find that a political regime change increases the probability of a growth acceleration by 5.3 percentage points while economic reforms are not related to growth accelerations.
We argue that these conclusions of Hausman et al. are wrong and the result of an error in their database. When we correct for this error and stick to the definition of a political regime change of Hausmann et al., we find that political regime changes are not related to the probability that a growth acceleration occurs. We also find some evidence that economic liberalization increases the probability of a growth acceleration, independent of whether this acceleration is sustained.
Our work can be seen as an illustration of the importance of replication as stressed by Hamermesh (2007). This paper contains a particular form of replication, namely redoing an analysis as published in a major journal using the data as used in that analysis to check whether the conclusions drawn are correct.^{ 1}
2. Our replication
For the period 19571992, Hausmann et al. (2005) identify no less than 83 periods of accelerated growth, using the following filter. For each country (with more than 1 million inhabitants and more than 20 available observations), the logarithm of real GDP per capita (taken from the Penn World Tables 6.2.) is regressed on time for every eight year period (n=7). That is,
Where y denotes real GDP capita and t is time. The estimated parameter, g_{t,t+n}_{ , }is taken as a proxy for the average growth rate over the period t to t+n and labeled the “least squares growth rate”._{ }To qualify as a growth acceleration, the least squares growth rate should be at least 3.5% per annum. Furthermore, it should be at least 2 percentage points higher than in the previous eight years. Finally, to rule out episodes of full economic recovery, the level of real GDP should be higher at the end of the acceleration than in all years before the acceleration. In cases that consecutive years qualify to be the start of a growth acceleration, the year is chosen with the highest Fstatistic of a piecewise linear (or spline) regression with the break at the relevant year. Hausmann et al. allow for the possibility that an acceleration is followed by another acceleration as long as the second acceleration starts at least five years after the first one.
We base our analysis on the definition and the identification of growth accelerations of Hausmann et al. (2005) – even though we feel that this definition can be improved upon – and focus on the explanatory variables used by these authors. These are categorized under three headings.

External shocks. Growth accelerations may be triggered by favorable external conditions and Hausmann et al. therefore include a terms of trade dummy, which takes the value 1 whenever the change in the terms of trade from year t4 to t is in the upper 10 percent of the entire sample.

Economic reform. To quantify a change in economic policy, the authors rely primarily on an index provided by Wacziarg and Welch (2003), that incorporates a number of structural features (e.g., presence of marketing boards and socialist economic regimes) and the macroeconomic environment (e.g., presence of a large blackmarket premium for foreign currency), in addition to tariff and nontariff barriers to trade. The variable included is a dummy that takes the value of 1 during the first five years of a transition towards “openness”.

Political regime changes are proxied by a dummy that takes a value of 1 in the 5year period beginning with a regime change as recorded in the Polity IV dataset, where a regime change is defined as either a threeunit change in the polity score or as a regime interruption.
Professor Rodrik kindly provided the data as used by Hausmann et al. (2005). We were able to reproduce their results (results available on request). However, in contrast to the definition given above, in the dataset of Hausmann et al. (2005) the political regime change dummy takes a value of 1 whenever there is a oneunit change in the Polity score. We have corrected this error and examine to what extent the results of Hausmann et al. change.
Table 1 shows the relationship between growth accelerations and regime changes (cf. Table 7 of Hausmann et al. 2005). We find that 21.7 percent of the accelerations are preceded or accompanied by a regime change, while Hausmann et al. report that around half of the growth accelerations are preceded or accompanied by regime changes. We also find that 14.2 (5) percent of the regime changes are followed by a (sustained) growth acceleration.^{2} The corresponding figures reported by Hausmann et al. are: 13.9 and 8.5 percent.
Table 1 here
Using the dataset as provided by Rodrik, we were able to fully reproduce Table 8 of Hausmann et al. (2005) in which they report on the relationship between the probability of a growth acceleration and a political regime change. As the first step in our subsequent analysis we corrected the coding mistake of political regime changes. Next, we checked the econometric specification of Hausmann et al. If we test for the restriction that all time dummies equal zero, it is not rejected for the model specifications as reported in columns (1)(9). Therefore, we omit the time dummies for those specifications, but include them in columns 10 and 11. Table 2 reports our results if we redo the regressions in Table 8 of Hausmann et al. using the corrected regime change variable and taking time dummies into account.^{3} Our results diverge substantially from those of Hausmann et al. Whereas the latter report that regime changes have a highly significant impact on the probability of the occurrence of a growth acceleration, our evidence suggests that, in general, regime changes are hardly related to growth accelerations. Take, for instance, column 10 in Table 2. Hausmann et al. find a coefficient of regime instability of 0.044 and a tstatistic of 4.16. If we correct for the error in the data set of Hausmann et al., we find instead a coefficient of 0.011 and a tstatistic of 0.87. Likewise, whereas Hausmann et al. find that the coefficients of positive and negative regime changes (Xposchange and Xnegchange, respectively) are generally highly significant, these variables are never significantly related to the probability of a growth acceleration according to our results. Furthermore, in our regressions the liberalization variable (Econ Lib) becomes significant at the 10 percent level, whereas Hausmann et al. find that this variable is always insignificant.^{4} The results for the other variables are similar to those of Hausmann et al.
Table 2 here
We also examined the determinants of sustained and unsustained growth accelerations. Table 3 reports our results when we estimate the models given in Table 12 of Hausmann et al. Again, we find that political regime changes are unrelated to growth accelerations. All other results are similar to the findings of Hausmann et al.
Table 3 here
3. Conclusions
We argue that the finding of Hausmann et al. (2005) that a political regime change increases the probability of an economic growth acceleration is wrong and the result of an error in their database. When we correct for this error and stick to the definition of political regime change of Hausmann et al., we find that regime changes do not affect the probability that a growth acceleration occurs. We also find some evidence that economic liberalization increases the probability of a growth acceleration, independent of whether this acceleration is sustained.
References
De Haan, Jakob (2007), “Political institutions and economic growth reconsidered”, Public Choice, 131 (3/4), 281292.
Jones, Benjamin F. and Benjamin A. Olken (2005), “The Anatomy of startstop growth”, NBER Working Paper, No. 11528.
Hamermesh, Daniel S. (2007), “Replication in Economics” NBER Working Paper No. 13026.
Hausman, Ricardo, Lant Pritchett and Dani Rodrik (2005), “Growth accelerations”, Journal of Economic Growth, 10, 303329.
Pritchett, Lant (2000), “Understanding patterns of economic growth: searching for hills among plateaus, mountains, and plains”, World Bank Economic Review, 14(2), 221150.
Wacziarg, Romain and Karen Horn Welch. (2003), “Trade Liberalization and Growth: New Evidence,” Stanford University (November).
Table 1. Regime changes and growth accelerations


Overlap of regime change and growth acceleration

# regime changes for which there is also accelerations data

127


# acceleration

83

18

# sustained accelerations

37

6

Table 2. Predicting growth accelerations (Dependent variable is a dummy for the timing of growth accelerations)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)













TOT_thresh90

0.039

0.040

0.040

0.037

0.035

0.032

0.032

0.033

0.038




(2.15)**

(2.21)**

(2.18)**

(2.10)**

(2.20)**

(2.10)**

(2.11)**

(2.15)**

(2.15)**



Econ Lib

0.039

0.043

0.040

0.040

0.007

0.014

0.014

0.014

0.040

0.031

0.032


(1.78)*

(1.93)*

(1.84)*

(1.83)*

(0.39)

(0.80)

(0.79)

(0.78)

(1.86)*

(1.55)

(1.59)

Xregchange

0.024









0.011



(1.66)*









(0.87)


Xposchange


0.001

0.000

0.001

0.002

0.000

0.000

0.000

0.002


0.007



(0.04)

(0.02)

(0.03)

(0.10)

(0.01)

(0.02)

(0.03)

(0.11)


(0.32)

Xnegchange


0.028

0.027

0.028

0.023

0.021

0.021

0.023

0.027


0.009



(1.22)

(1.17)

(1.20)

(1.21)

(1.18)

(1.16)

(1.25)

(1.19)


(0.42)

Leader Death



0.030

0.062

0.000

0.000

0.001

0.001

0.062






(1.20)

(1.94)*

(0.00)

(0.01)

(0.01)

(0.02)

(1.92)*



Tenure




0.006

0.045

0.048

0.048

0.048

0.006







(1.98)**

(3.06)***

(3.00)***

(3.00)***

(3.00)***

(1.95)*



War End







0.002

0.010

0.027










(0.16)

(0.59)

(1.28)



Civil War








0.019

0.017











(0.82)

(0.64)



Finance





0.035

0.114

0.113

0.116









(1.64)

(2.85)***

(2.85)***

(2.90)***




Finance Dev






0.044

0.044

0.044










(2.28)**

(2.28)**

(2.34)**




Observations

2094

2094

2094

2094

1891

1891

1891

1891

2094

2739

2739

Pseudo Rsquared

0.01

0.01

0.01

0.01

0.02

0.02

0.02

0.02

0.02

0.04

0.04

Time dummies equal 0, prob> Chi^2

0.7667

0.7670

0.7725

0.6986

0.5015

0.5648

0.5519

0.561

0.6861

0.0048

0.005

Time dummies included

No

No

No

No

No

No

No

No

No

Yes

Yes

Robust z statistics in parentheses












* significant at 10%; ** significant at 5%; *** significant at 1%










Table 3. Predicting sustained and unsustained growth accelerations
(Dependent variable is a dummy for the timing of growth accelerations)

(1)

(2)

(3)

(4)

(5)

(6)


All

All

Sustained

Sustained

Sustained

Unsustained

TOT_thresh90

0.067

0.074

0.019

0.016


0.023


(2.80)***

(3.25)***

(1.35)

(1.22)


(3.67)***

Econ Lib

0.090

0.093


0.140

0.091



(1.98)**

(2.08)**


(3.94)***

(3.50)***


Xposchange

0.012

0.008

0.010

0.015

0.013

0.002


(0.33)

(0.23)

(0.44)

(0.66)

(0.65)

(0.40)

Xnegchange

0.046

0.050

0.017

0.019

0.011

0.005


(1.76)*

(1.92)*

(1.04)

(1.18)

(0.63)

(1.20)

Finance

0.006





0.994


(0.17)





(8.15)***

Observations

1211

1300

1300

1300

1723

1140

Pseudo Rsquared

0.02

0.03

0.01

0.05

0.12

0.11

Time dummies equal 0, prob > Chi^2

0.2254

0.3175

0.9793

0.9794

0.0001

0.0000

Time dummies included

No

No

No

No

Yes

Yes

Robust z statistics in parentheses







* significant at 10%; ** significant at 5%; *** significant at 1%





